Drivers of 30-day Readmission, 5-year Survival and Length of Stay Following the Treatment of Brain Cavernous Angiomas: An Analysis of the National Cancer Database (NCDB) Using Machine Learning

Introduction: Cerebral cavernous angiomas (CCAs) are dilated blood vessels in a venous–capillary vascular bed without brain parenchyma. Although surgical resection is preferred; targeted radiotherapy or chemotherapy can be applied in some cases. In our study, we aim to investigate treatment modalities in patients diagnosed with CCAs and to determine the drivers impacting overall 30-day readmission, long-term survival and postoperative length of stay (LOS).

Methods: We queried the National Cancer Database between 2004 and 2016 for patients diagnosed with CCAs. Demographic, clinical, and outcomes data were compiled. Five-year survival for each treatment was compared using the Kaplan-Meier and Nelson-Aalen curves with statistical comparisons based on the log-rank test. Multivariate analysis was performed to find the drivers of survival, readmission, and LOS. Supervised machine learning (ML) models were trained to predict survival, and Shapley additive explanations (SHAP) were calculated using the best-performing model for survival’s feature importance.

Results: We found 1,737 patients diagnosed and treated with cavernous brain angiomas. The most common treatment was surgical resection (N=1,504), followed by chemotherapy (N=155), then radiotherapy (N=78). Bilaterally located tumors, and surgical resection alone (p < 0.05), were more likely to predict 30-day readmission. Age, male sex, high comorbidity score, large ventricular lesion, tumor size, and chemotherapy alone are more likely to predict mortality (p < 0.05). Brain stem lesions are more likely to affect LOS following resection (p < 0.01). The best-performing ML model predicting mortality was Random Forest (97.3%). Increased odds of survival are associated with surgery alone compared to radiotherapy alone or chemotherapy alone (p < 0.005). SHAP showed that age, surgery alone, insurance status, and laterality impacted mostly long-term survival.

Conclusion : Our analysis showed that bilaterally located CCAs are more likely to increase the risk of 30-day readmission; large ventricular lesions are more likely to increase mortality risk,while brain stem lesions are more likely to increase LOS following treatment.